We propose a novel trajectory representation learning approach that efficiently learns travel semantics, including movement patterns and travel purposes, from vehicle trajectories.